BusinessOptix MCP
BusinessOptix MCP makes a BusinessOptix library available to an AI assistant, so that people can ask questions about their processes, models and associated data in plain language and get answers drawn from that library rather than from the assistant's general knowledge.
It is built on the Model Context Protocol (MCP), an open standard for connecting AI assistants to the systems where an organisation's information actually lives. Any MCP-capable assistant can connect. BusinessOptix has tested with Claude and with the MCP Inspector, a reference tool for exercising MCP servers directly.
The practical effect is that the process knowledge held in BusinessOptix becomes available where people are already working and asking questions, without them having to know which model to open or which report to run.
What it can do
An enabled library offers three capabilities. The assistant chooses between them according to what is being asked — nobody selects a capability, they just ask.
Find models. The catalogue of models in the library: what exists, and how to identify it. It can be narrowed by name, type, author, owner, status, version and dates.
Search the library. A query across the library's content, underpinned by the same library-wide search used in the BusinessOptix application. It reaches every kind of content the library holds, and can be narrowed by the roles, systems and locations referenced in that content.
Answer questions. Structured responses drawn from BusinessOptix data sources — the platform's functional wrapper around the data held against models, turning model content into a defined set of rows and columns. Answers can be filtered, sorted, grouped and totalled, so questions like how many, which of these and broken down by are answered from the data rather than estimated.
How it behaves
Four characteristics shape how BusinessOptix MCP fits into an organisation.
It is read-only. It answers questions. It does not create, edit or delete anything. Connecting an assistant cannot change library content, and the permissions granted at connection include no ability to do so.
It acts as the person using it. Every request is authorised on behalf of the signed-in user, using their own permissions. An assistant sees exactly what that person would see signing in to BusinessOptix directly — no more. Content outside their permissions is not returned, and is not counted, summarised or otherwise implied. Where a data source enforces its own access control, that applies too. Two people asking the same question of the same library may correctly receive different answers.
It covers one library at a time. A connection reaches a single library and nothing else. Each enabled library has its own address, and an assistant is connected to each one separately, so connecting to one library does not extend access to any other.
It is enabled deliberately. MCP is switched on per library, at the customer's request. Nothing is exposed by default, and both the customer and BusinessOptix retain control over which libraries are reachable.
Access is authorised for 14 days at a time, after which the user signs in and approves again, and every request is logged for administrators to review.
Where BusinessOptix's responsibility ends
The assistant is not supplied by BusinessOptix. You choose it — Claude, Copilot, Gemini or another MCP-capable assistant — and you hold the relationship with its provider.
BusinessOptix controls what leaves the library: only content the signed-in user is entitled to see, only in response to what they ask, and only to the assistant they have explicitly authorised. What that provider then does with the content it receives — retention, processing location, whether it is used for training — is governed by your own agreement with them, not by BusinessOptix.
So the question "where does our process content go?" has two halves. BusinessOptix answers the first: nothing leaves except in response to an authorised user's question, and every request is logged. The second belongs to the terms you hold with your assistant provider, and is worth confirming before connecting a library.
Useful immediately, and sharper over time
The capabilities become available at different points.
Finding models and searching the library work the moment MCP is enabled. They reach everything in the library the person asking can already see, with no configuration, no curation and nothing to prepare.
Answering questions draws on the data sources you choose to expose. A library can hold hundreds of data sources, and an author decides which of them an assistant may draw on. Each one enabled and described adds to the set of questions the library can answer precisely.
So a library is useful for finding and reading from the day it is switched on, and grows into answering exact questions as its data sources are curated.
The set of questions is not fixed, and not limited to what BusinessOptix supplies. BusinessOptix provides some data sources, customers build their own, and implementation projects often add more — so a library's answers can be shaped around the questions that organisation actually asks.
What it takes
| Who does it | Article | |
|---|---|---|
| Enable MCP support on the library, and install the MCP for Data Sources extension | BusinessOptix customer support | Enabling MCP on a library |
| Connect an assistant and start asking | Anyone with a BusinessOptix account | Using BusinessOptix MCP |
| Make the library answer precise questions | An author | Making your library answer questions |
Enablement has two parts, and both are worth requesting together: the first makes the library reachable, and the second is what allows data sources to be exposed as questions.
Using MCP works on any licence type. Curating data sources for it needs an author licence.
For technical readers. The three capabilities are delivered as five MCP tools, exposed by a server and invoked by an MCP client:
model_listandmodel_search, thenlist_questions,describe_questionandask_questionfor answers. The server exposes no MCP resources or prompts. Tool selection is the assistant's decision, informed by the tool and data source descriptions the server publishes — which is why those descriptions matter. Authorisation is OAuth; bearer tokens are also supported for non-interactive use.